Research On Edge Computing Task Scheduling Strategy For LEO Constellation Broadband Network

Deploying edge computing servers in low Earth orbit(LEO)satellite broadband networks eliminates the need for satellite relay of computing tasks to ground cloud centers,effectively reducing task response latency,minimizing bandwidth consumption,and providing users with more secure and real-time services.Task scheduling is a key technology in edge computing.Traditional task scheduling strategies are mostly designed for ground networks and are not fully applicable to the constantly changing topology of LEO satellite broadband networks.Therefore,research on task scheduling strategies for edge computing in LEO satellite broadband networks has important theoretical and practical significance.This Thesis aims to reduce latency and energy consumption and proposes a multitask LEO satellite broadband network edge computing task scheduling strategy.The research work is mainly divided into three aspects.Firstly,a multi-level task scheduling architecture based on inter-satellite collaboration for LEO satellite broadband network edge computing is proposed,detailing elements such as user terminals,LEO satellite broadband networks,ground cloud centers,and links,and designing a task scheduling process.Secondly,more general partial scheduling strategies in task scheduling are studied,and tasks are comprehensively modeled,including multi-task models,subtask models,task response latency models,and task energy consumption models.Finally,a LEO satellite broadband network edge computing task scheduling strategy is proposed.To solve the dependencies between subtasks in partial scheduling,a priority-based subtask selection strategy(PSS)is designed.PSS is a preprocessing of tasks,and its output result is the input of the scheduling strategy.The edge computing task scheduling decision problem in LEO satellite broadband networks is modeled as a binary integer programming problem.Since the problem is NP-hard,the LEO satellite broadband network scenario is complex,and a single task may be completed by multiple edge computing nodes cooperating,the task scheduling problem is transformed into a partially observable Markov decision process.With satellite nodes as agents,action spaces,observation spaces,global state spaces,and reward functions are designed according to task modeling.Based on this,a partial scheduling strategy based on multi-agent reinforcement learning(MAPSS)is proposed,adopting centralized training and decentralized execution methods.Each satellite adaptively makes scheduling decisions,and the relationship between them is fully cooperative to achieve optimal scheduling decisions.This thesis uses the Iridium satellite constellation as the target simulation scenario,and evaluates the performance of MAPSS in terms of strategy characteristics and strategy performance under different objectives.In strategy characteristics,convergence analysis and scheduling hop count analysis are conducted.The simulation results show that MAPSS can effectively converge in LEO satellite broadband network scenarios,and more than 98% of tasks can be processed within three hops,proving that MAPSS can limit the state space to a smaller range.In addition,comparative simulations are conducted for three objectives: maximizing average response latency gain,maximizing average energy consumption gain,and maximizing average task gain.The simulation results show that MAPSS performs well as the scale of subtasks and the number of users increase,demonstrating that MAPSS can effectively reduce latency and energy consumption and adapt well to different types of task requirements.